Efficient learning-based blur removal method based on sparse optimization for image restoration

Abstract

In imaging systems, image blurs are a major source of degradation. This paper proposes a parameter estimation technique for linear motion blur, defocus blur, and atmospheric turbulence blur, and a nonlinear deconvolution algorithm based on sparse representation. Most blur removal techniques use image priors to estimate the point spread function (PSF); however, many common forms of image priors are unable to exploit local image information fully. In this paper, the proposed method does not require models of image priors. Further, it is capable of estimating the PSF accurately from a single input image. First, a blur feature in the image gradient domain is introduced, which has a positive correlation with the degree of blur. Next, the parameters for each blur type are estimated by a learning-based method using a general regression neural network. Finally, image restoration is performed using a half-quadratic optimization algorithm. Evaluation tests confirmed that the proposed method outperforms other similar methods and is suitable for dealing with motion blur in real-life applications. OPEN ACCESSCitation: Yang H, Su X, Chen S, Zhu W, Ju C (2020) Efficient learning-based blur removal method based on sparse optimization for image restoration. PLoS ONE 15(3): e0230619. https:// introduced a sparsity function with an 0 -norm constraint, while Pan et al. [6] estimated the blur kernel from the dark channel of the blurry images. These estimation techniques perform well when dealing with hand-drawn PSFs. However, in real-life situations, the blur models are often known. For example, when monitoring targets on a conveyor belt with a fixed camera, the PSF can be modeled based on the motion length during the exposure time. Similarly, with respect to space exploration, the atmospheric turbulence can be modeled by a Gaussian function, whose variance indicates the blur degree. Defocus blur can be modeled based on the defocus radius. It is more convenient and practical to solve a parameter identification problem than to estimate the PSF. Given this fact, Jalobeanu et al. [7] used the maximum likelihood estimator (MLE) on the entire dataset available to estimate the parameters for a Gaussian model. Yin and Hussain [8] combined the non-Gaussianity measures for independent component analysis to estimate the parameters for blur models. Dash and Majhi [9] suggested a radial basis function neural network with image features based on the magnitude of Fourier coefficients to estimate the motion lengths. Yan and Shao [10] proposed a supervised deep neural network to classify the blur type and adopted the expected patch log likelihood method [11] to restore the latent image. Further, Kumar et al. [12] used the Tchebycheff moment to estimate the Gaussian variance for turbulence blurs.With a known PSF, the latent image can be restored using inverse filters or some other nonblind deconvolution method. Levin et al. [13] proposed a hyper-Laplacian prior and adopted the iterative reweighted least squares (IRLS) algorithm to solve the optimi...

X-ray micro-tomographic data of live larvae of the beetle Cacosceles newmannii

“…Thus, when using hard cut-offs in tracheal diameter, it is difficult to know how much of total tracheal volume that actually is quantified, and how much is found in trachea and tracheoles with smaller diameters. Future studies could explore possibilities to improve minimum resolution, for instance using machine learning algorithms that improve capacity to distinguish tracheal structures in fuzzy images (Yang et al, 2020). More generally, it would be important to study insect species across a wide range of body sizes to determine the distribution of tracheal size relationships (Aitkenhead et al, 2020;Kaiser et al, 2007) at different scanning resolutions (Iwan et al, 2015).…”

X-ray micro-tomographic data of live larvae of the beetle Cacosceles newmannii

“…Thus, when using hard cut-offs in tracheal diameter, it is difficult to know how much of total tracheal volume that actually is quantified, and how much is found in trachea and tracheoles with smaller diameters. Future studies could explore possibilities to improve minimum resolution, for instance using machine learning algorithms that improve capacity to distinguish tracheal structures in fuzzy images [ 5 ] . More generally, it would be important to study insect species across a wide range of body sizes to determine the distribution of tracheal size relationships [ 6 , 21 ] at different scanning resolutions [ 14 ] .…”

Improved estimation of motion blur parameters for restoration from a single image

“…However, these efficient infrastructures are not suitable to the application of image deblurring, since the research theory of image restoration is different from other image processing areas. A popular way to tackle motion deblurring of single image is to deconvolute the blurred image with PSF [19].…”